Section 11

LTE
Basic treatments of stellar atmospheres adopt, as a starting point, the assumptions of Local

Thermodynamic Equilibrium (LTE), and hydrostatic equilibrium. The former deals with the

microscopic properties of the atoms, and we will discuss it here; the latter addresses the

large-scale conditions (and applies throughout a normal star), and is discussed in Section 10.1.

11.1 Local Thermodynamic Equilibrium

Fairly obviously, in Local Thermodynamic Equilibrium (LTE) it is assumed that all

thermodynamic properties in a small volume have the thermodynamic equilibrium values at the

local values of temperature and pressure.

Specically, this applies to quantities such as the occupation numbers of atoms, the opacity,

emissivity, etc. The LTE assumption is equivalent to stating that

1. the electron and ion velocity distributions are Maxwellian,

3/2
−mv 2
  
dn(v) m
=n exp
dv 2πkTk 2kTk

for number density n of particles of mass m at kinetic temperature Tk ;

2. the photon source function is given by the Planck function at the local temperature (i.e.,

Sν = Bν , and jν = kν Bν ).

3. the excitation equilibrium is given by the Boltzmann equation

 
nj gj −(Ej − Ei )
= exp (11.1)
ni gi kT

103
 4. the ionization equilibrium is given by the Saha equation

(2πme kT )3/2
 
ne n2,1 2g2,1 −χ1,i
= exp (11.2)
n1,i g1,i kT h3
where 1, i, 2, 1 denote levels i, 1 in ionization stages 1, 2.

One might augment this list with the perfect-gas equation of state, P = nkT , but since this

applies under many circumstances where LTE doesn't hold, it's not usually mentioned in this

context.

If a process is purely collisional, conditions are, naturally, determined on a purely local basis

locally, and LTE applies. We have already encountered one such situation where LTE is a good

approximation: free-free emission results from a purely collisional process, justifying our

adoption of Sν = Bν (Section 6).

If radiation plays a role, then provided the photon and particle mean free paths are short

compared to the length scales over which conditions change signicantly (i.e., if the opacity is

high), then we can again expect LTE to be a reasonable assumption; this is a good

approximation in stellar interiors.

In stellar atmospheres the LTE approximation may be a poor one, as photon mean free paths

are typically larger than those of particles. Thus one region can be aected by the radiation

eld in another part of the atmosphere (e.g., a deeper, hotter region). As a rule of thumb,

therefore, LTE is a poor approximation if the radiation eld is important in establishing the

ionization and excitation equilibria (as in hot stars, for example). It's more likely to be

acceptable when particle densities are high and the radiation eld is relatively weak; for stars,

this means higher gravities (i.e., main-sequence stars rather than supergiants) and cooler

eective temperatures. When LTE breaks down, we have a `non-LTE' (nLTE) situation, and

level populations must be calculated assuming statistical equilibrium (section 2.3.2).

11.2 The Saha Equation

The Boltzmann Equation gives the relative populations of two bound levels i and j, in some

initial (or `parent') ionization stage `1':

 
n1,j g1,j −(E1,j − E1,i )
= exp (11.1)
n1,i g1,i kT

where E1,i & E1,j are the level energies (measured from the ground state, E1,1 = 0), and g1,i &

g1,j are their statistical weights (2J + 1, where J is the total angular-momentum quantum

number).

104
To generalize the Boltzmann eqtn. to deal with collisional ionization to the next higher (or

`daughter') ionization stage `2', we identify the upper level j with a continuum state; n1,j , the
number of parent ions in excitation state j, then equates with n2,1 (v), the number of ionized
atoms where the detached electron has velocity v.

(Note that ionization stages `1' and `2' always represent any two consecutive stages  for
0 +
example, H and H , or C
2+ and C3+ .)

The total statistical weight of the ionized system is given by the combined statistical weights1 of
the newly created ion and the electron, i.e., g2 ge (v); while the relevant energy is the sum of the
ionization energy and the kinetic energy of the free electron. Thus we have
( )
n2,1 (v) g2 ge (v) −(χ1,i + 21 me v 2 )
= exp (11.3)
n1,i g1,i kT

where χ1,i = E∞ − E1,i is the ionization potential for level i in the parent species.
An aside: The statistical weight of a free electron. The statistical weight of a free electron

is just the probability of nding it in a specic cell of `phase space'. Since the state of a free
particle is specied by three spatial coördinates x, y, z and three momentum coördinates
p(x), p(y), p(z), the number of quantum states (for which the statistical weights are each 1) in an
element of phase space,

dx dy dz · dp(x) dp(y) dp(z) = dN

is given by
2 dN 2
ge (v) = = 3 dx dy dz · dp(x) dp(y) dp(z)
h3 h
where h is Planck's constant and the factor 2 arises because the electron has two possible spin
states. The statistical weight per unit volume is thus
2
dp(x) dp(y) dp(z)
h3
for a single electron. However, there may be other free electrons, from other ions, which occupy
some of the available states in the element of phase space dN . If the number density of electrons is
ne then the eective volume available to a collisionally ejected electron is reduced by a factor 1/ne .
Thus the statistical weight available to a single free electron is
2 dp(x) dp(y) dp(z)
ge (v) = .
ne h3
Furthermore, if the velocity eld is isotropic, the `momentum volume' can be replaced simply by
its counterpart in spherical coördinates,

dp(x) dp(y) dp(z) = 4πp2 dp

Using these results we can write eqtn. (11.3) as

( ` ´)
ne n2,1 (v) 2gi − χ1,i + 12 me v 2
= 3 exp 4πp2 dp
n1,i h g1,i kT

1
The statistical weight is a form of probability, and the probability of `A and B', P (A+B), is the product
P (A)P (B).

105
 but the momentum p = me v ; i.e., 12 me v 2 = p2 /(2me ), whence

−p2
n −χ o  ff
ne n2,1 (v) 2g2 1,i
= 3 exp exp 4πp2 dp.
n1,i h g1,i kT 2me kT

Since we're interested in the ionization balance (not the velocity distribution of the ionized
electrons), we integrate over velocity to obtain the total number of daughter ions:
Z∞
−p2
n −χ o  ff
ne n2,1 2g2 1,i
= 3 exp 4π p2 exp dp.
n1,i h g1,i kT 2me kT
0

We can then use result of a standard integral,

Z∞
√ `
x2 exp −a2 x2 dx = π/ 4a3
` ´ ´

to obtain

(2πme kT )3/2
 
ne n2,1 2g2 −χ1,i
= exp (11.2)
n1,i g1,i kT h3

This is one common form of the Saha Equation (often expressed in terms of the ground state of

the parent ion, n1,1 ).

11.3 Partition functions

The version of the Saha equation given in eqtn. (11.2) relates populations in single states of

excitation for each ion. Generally, we are more interested in the ratios of number densities of

dierent ions summed over all states of excitation  i.e., the overall ionization balance. We

determine this by dening the partition function as

X
U= gn exp (−En /kT )
n

(an easily evaluated function of T ), whence

2U2 (2πme kT )3/2

 
ne n2 −χ1
= exp (11.4)
n1 U1 h3 kT

where we use χ1 , the ground-state ionization potential of the parent atom, as it is to this that

the partition function is referred (i.e., E1,1 ≡ 0).

Since the electron pressure is Pe = ne kT we can also express the Saha equation in the form

2U2 (2πme )3/2 (kT )5/2

 
n2 −χ1
= exp (11.5)
n1 U1 h3 Pe kT

106
11.3.1 An illustration: hydrogen

The Balmer lines of hydrogen, widely observed as absorption lines in stellar spectra, arise

through photoexcitation from the n=2 n = 2 level,

level of neutral hydrogen. To populate the
5
we might suppose that we need temperatures such that kT ' E1,2 = 10.2eV; i.e., T ' 10 K.
4
However, the Hα line strength peaks in A0 stars, which are much cooler than this (T ∼ 10 K).

Why? Because we need to consider ionization as well as excitation. We therefore need to

combine the Saha and Boltzmann equations to obtain the density of atoms in a given state of

excitation, for a given state of ionization.

We express the Boltzmann equation, eqtn. (11.1), in terms of the partition function U:
 
n1,2 g1,2 −E1,2
= exp
n1 U1 kT

where n1 is the number density of H

0 atoms in all excitation states and E1,2 is the excitation

energy of the n=2 level (10.2 eV); that is,

 
g1,2 −E1,2
n1,2 = exp n1 .
U1 kT

However, the total number of hydrogen nuclei is n(H) = n1 + n2 = n1 (1 + n2/n1 ); that is,

n1 = n(H)(1 + n2/n1 )−1 . Using this, and n2/n1 from eqtn. (11.4), we nd
" #!−1
2U2 (2πme kT )3/2
  
g1,2 −E1,2 −χ1
n1,2 = exp 1+ exp n(H)
U1 kT n1 U1 h3 kT
" #!−1
2U2 (2πme )3/2 (kT )5/2
  
g1,2 −E1,2 −χ1
= exp 1+ exp n(H)
U1 kT n1 U1 h3 Pe kT

We can now see why the Balmer lines peak around 104 K: while higher temperatures give larger
populations n1,2 /n(H0 ), they give smaller
0
populations n(H )/n(H). The overall result is that

n1,2 /n(H) peaks around 10kK.

The Saha equation also gives an explanation of why supergiant stars are cooler than

main-sequence stars of the same spectral type. Spectral characteristics are dened by ratios of

lines strengths; e.g., O-star subtypes are dened by the ratio (He ii λ4542)/(He i λ4471), which
+ 0
in turn traces the ratio He /He . Of course, higher temperatures increase the latter ratio.

However, a supergiant star has a lower surface gravity (and atmospheric pressure) than a

main-sequence star. From eqtn. (11.5) we see that a lower pressure at the same temperature

gives rise to a larger ratio n2 /n1 , so for two stars of the same temperature, the supergiant has

an earlier spectral type (or, equivalently, at the same spectral type the supergiant is cooler).

107
108
Section 12

Stellar Timescales
12.1 Dynamical timescale

12.1.1 `Hydrostatic equilibrium' approach

If we look at the Sun in detail, we see that there is vigorous convection in the envelope. With

gas moving around, is the assumption of hydrostatic equilibrium justied? To address this

question, we need to know how quickly displacements are restored; if this happens quickly

(compared to the displacement timescales), then hydrostatic equilibrium remains a reasonable

approximation even under dynamical conditions.

We have written an appropriate equation of motion,

dP
ρa = ρg + (10.2)
dr

where g is the acceleration due to gravity and

2
d r
a=
dt 2

is the nett acceleration. As the limiting case we can `take away' gas-pressure support (i.e., set

dP /dr = 0), so our equation of motion for collapse under gravity is just

2
d r Gm(r)
=− .
dt 2 r2

Integrating (and taking r from the surface inwards),

Gm(r) t2 1
r= = gt2 (for initial velocity v0 = 0). (12.1)
r2 2 2

109
Identifying the time t in eqtn. (12.1) with a dynamical timescale, we have

s
2r3
tdyn = . (12.2)
Gm(r)

Departures from hydrostatic equilibrium are restored on this timescale (by gravity in the case

of expansion, or pressure in the case of contraction). In the case of the Sun,

s
2R3
tdyn = ' 37 min.
GM

If you removed gas-pressure support from the Sun, this is how long it would take a particle at

the surface to free-fall to the centre. Since the geological record shows that the Sun hasn't

changed substantially for at least 109 yr, it is clear that any departures from hydrostatic

equilibrium must be extremely small on a global scale.

We might expect departures from spherical symmetry to be restored on a dynamical timescale (in
the absence of signicant centrifugal forces), and by indeed comparing eqtns. (10.6) and (12.2) we
see that spherical symmetry is appropriate if
√
2
ω
tdyn

12.1.2 `Virial' approach

The `hydrostatic equilibrium' approach establishes a collapse timescale for a particle to fall

from the surface to the centre. As an alternative, we can consider a timescale for gas pressure

to ll a void  a pressure-support timescale. Noting that a pressure wave propagates at the

sound speed, this dynamical timescale can be equated to a sound-crossing time for transmitting

a signal from the centre to the surface.

The sound speed is given by

 
kT
c2S =γ (12.3)
µm(H)

(where γ = Cp /Cv , the ratio of specic heats at constant pressure and constant volume).

From the virial theorem,

2U + Ω = 0 (10.11)

with
Z Z
3 3 ρ(r)
U= kT n(r) dV = kT dV (10.8)
V 2 V 2 µm(H)

110
and

Z M Z
Gm(r) Gm(r)
Ω=− dm = − ρ(r) dV (10.10)
0 r V r

so that

3kT Gm(r)
= ; (12.4)
µm(H) r

that is, from eqtn. (12.3),

3 2 Gm(r)
cS = (12.5)
γ r

For a monatomic gas we have γ = 5/3, giving, from eqtns. (12.3) and (12.4),

5 Gm(r)
c2S =
9 r
so that the (centre-to-surface) sound crossing time is

s
r 9/5r 3
t= = (12.6)
cS Gm(r)

(which is within ∼10% of eqtn. (12.2)).

12.2 Kelvin-Helmholtz and Thermal Timescales

Before nuclear fusion was understood, the conversion of potential to radiant energy, through

gravitational contraction, was considered as a possible source of the Sun's luminosity.

1 The

time over which the Sun's luminosity can be powered by this mechanism is the

Kelvin-Helmholtz timescale.

The available gravitational potential energy is

M
−Gm(r)
Z
Ω= dm (10.10)
0 r

but

4
m(r) = πr3 ρ so dm = 4πr2 ρ dr
3
1
Recall from Section 10.4 that half the gravitational potential energy lost in contraction is radiated away, with
the remainder going into heating the star.

111
and

R
16π 2 4 2
Z
Ω= −G r ρ (r) dr
0 3
16 2 2 5
'− π Gρ R (12.7)
15
(assuming ρ(r) = ρ(R)).

The Kelvin-Helmholtz timescale for the Sun is therefore

|Ω()|
tKH = . (12.8)
L

which for ρ = 1.4 × 103 kg m

−3 , Ω = 2.2 × 1041 J is tKH ' 107 2
yr.

The Kelvin-Helmholtz timescale is often identied with the thermal timescale, but the latter is

more properly dened as

U ()
tth = , (12.9)
L

which (from the virial theorem) is ∼ 1/2tKH . In practice, the factor 2 dierence is of little

importance for these order-of-magnitude timescales.

12.3 Nuclear timescale

We now know that the source of the Sun's energy is nuclear fusion, and we can calculate a

corresponding nuclear timescale,

f M c2
tN = (12.10)
L
where f is just the fraction of the rest mass available to the relevant nuclear process. In the

case of hydrogen burning this fractional `mass defect' is 0.007, so we might expect

0.007M c2
tN = (' 1011 yr for the Sun).
L

However, in practice, only the core of the Sun  about ∼10% of its mass  takes part in

hydrogen burning, so its nuclear timescale for hydrogen burning is ∼ 1010 yr. Other

evolutionary stages have their respective (shorter) timescales.

2
At the time that this estimate was made, the fossil record already indicated a much older Earth (∼ 109 yr).
Kelvin noted this discrepancy, but instead of rejecting contraction as the source of the Sun's energy, he instead
chose to reject the fossil record as an indicator of age.

112
12.4 Diusion timescale for radiative transport

Deep inside stars the radiation eld is very close to black body. For a black-body distribution

the average photon energy is

E = U/n ' 4 × 10−23 T [J photon

−1
]. (1.28)

The core temperature of the Sun is Tc ' 11/2 × 107 K (cf. eqtn. 10.17), whence E = 3.5 keV 

i.e., photon energies are in the X-ray regime.

Light escaping the surface of the Sun (Teff ' 5770K) has a mean photon energy ∼ 3 × 103
smaller, in the optical.

The source of this degradation in the mean energy is the coupling between radiation and

matter. Photons obviously don't ow directly out from the core, but rather they diuse
through the star, travelling a distance of order the local mean free path, `, before being

absorbed and re-emitted in some other direction (a `random walk'). The mean free path

depends on the opacity of the gas:

` = 1/kν = 1/(κν ρ) (2.4)

where kν is the volume opacity (units of area per unit volume, or m

−1 ) and κν is the mass
2
opacity (units of m kg
−1 ).

√
After nsc scatterings the radial distance travelled is, on average, nsc ` (it's a statistical,

random-walk process). Thus to travel a distance R we require

 2
R
nsc = . (12.11)
`
Solar-structure models give an average mean free path ` ' 1 mm (incidentally, justifying the

LTE approximation in stellar interiors); with R ' 7 × 108 m,

nsc ' 5 × 1023

The total distance travelled by a (ctitious!) photon travelling from the centre to the surface is

nsc × ` ' 5 × 1020 m (∼ 1012 R !), and the time to diuse to the surface is

(nsc × `)/c ' 5 × 104 yr.

[More detailed calculations give 17 × 104 yr; why? Naturally, there are regions within the Sun

that have greater or lesser opacity than the average value, with the largest opacities in the

central ∼0.4R and in the region immediately below the photosphere. Because of the `square

root' nature of the diusion, a region with twice the opacity takes four times longer to pass

through, while a region with half the opacity takes only 1.414 times shorter; so any

non-uniformity in the opacity inevitably leads to a longer total diusion time.]

113